I've read this twice and have some questions:

1. What's the context/goal of flagging relationship milestones?

2. What does a normal relationship milestone look like?

3. Who's the intended audience of this case study? (The prior 2 questions may be moot because I don't have the right background information.)

I’ve also noticed how powerful actual keywords are when trying to retrieve data from a sea of candidates that have high cosine similarity to the query. My quick and dirty ranking mechanism is simply keyword match first, and only then similarity score.